Intelligent page store for concurrent transaction and query processing

Abstract

An intelligent page store is proposed that provides two paths for accessing data. Update and transaction traffic are presented the current version of the data by the page store. Complex queries against the same data are presented a recent transaction consistent snapshot of the data. Periodically, new snapshots are created for access by new queries. By using an implicit versioning scheme, the page store maintains one copy of most of the data, while providing consistent access. This scheme allows the transaction and query systems to be independently optimized. We show that transaction performance is largely unaffected by concurrent queries, and disks are used efficiently since a query copy of the database is obviated.